Change deprecated name to has_offloaded_params (vllm-project#1556)

The `is_module_offloaded` util is now deprecated and has been
implemented under the name `has_offloaded_params`.
`has_offloaded_params` is the name adopted by accelerate.


https://github.com/neuralmagic/compressed-tensors/blob/main/src/compressed_tensors/utils/offload.py#L118-L120

Signed-off-by: Kyle Sayers <[email protected]>
Co-authored-by: Dipika Sikka <[email protected]>

Author: kylesayrs

src/llmcompressor/transformers/sparsification/compressed_tensors_utils.py

@@ -10,7 +10,7 @@
     ModelCompressor,
     SparsityCompressionConfig,
     delete_offload_parameter,
-    is_module_offloaded,
+    has_offloaded_params,
     register_offload_parameter,
 )
 from loguru import logger
@@ -138,7 +138,7 @@ def untie_word_embeddings(model: PreTrainedModel):
             continue
 
         # this could be replaced by a `get_offloaded_parameter` util
-        if not is_module_offloaded(module):
+        if not has_offloaded_params(module):
             untied_data = module.weight.data.clone()
         else:
             untied_data = module._hf_hook.weights_map["weight"].clone()